Publication number | US7584455 B2 |
Publication type | Grant |
Application number | US 10/844,710 |
Publication date | Sep 1, 2009 |
Filing date | May 12, 2004 |
Priority date | Oct 23, 2003 |
Fee status | Lapsed |
Also published as | US20050229044 |
Publication number | 10844710, 844710, US 7584455 B2, US 7584455B2, US-B2-7584455, US7584455 B2, US7584455B2 |
Inventors | Thomas J. Ball |
Original Assignee | Microsoft Corporation |
Export Citation | BiBTeX, EndNote, RefMan |
Patent Citations (20), Non-Patent Citations (41), Referenced by (79), Classifications (8), Legal Events (6) | |
External Links: USPTO, USPTO Assignment, Espacenet | |
This application claims the benefit of U.S. Provisional Patent Application No. 60/557,034, entitled “Predicate-Complete Test Coverage and Generation,” filed Mar. 25, 2004, and U.S. Provisional Patent Application No. 60/513,887, entitled, “Abstraction-Guided Test Generation: A Case Study,” filed Oct. 23, 2003, which are incorporated herein by reference.
This application relates to computer program testing, and more particularly relates to test coverage and test generation in computer program testing.
In the field of computer program testing, different approaches have been developed to more accurately and completely test program function. For example, control-flow-based test generation seeks to cover at a minimum all the statements or branches in a program (“statement coverage” or “branch coverage”). In multiple condition coverage, each possible condition of each statement is tested at a local level. For example, in a statement of the form if (x∥y) then a else b, multiple condition coverage would locally test whether the statement performs as expected where x is true and y is false, where y is true and x is false, where both x and y are true, or where both x and y are false. Path coverage is an extension of multiple condition coverage that seeks to test all possible paths through a program. The idea of using paths and symbolic execution of paths to generate tests has a long and rich history going back to the mid-1970's and continuing to the present day. See, e.g., Boyer et al., “SELECT—A Formal System for Testing and Debugging Programs by Symbolic Execution,” SIGPLAN Notices, 10(6):234-245 (1975); Howden, “Reliability of the Path Analysis Testing Strategy,” IEEE Trans. on Software Engineering, 2:208-215 (1976); Clarke, “A System to Generate Test Data and Symbolically Execute Programs,” IEEE Trans. on Software Engineering, 2(3):215-222 (1976); Jasper et al., “Test Data Generation and Feasible Path Analysis,” Proc. Int'l Symposium on Software Testing and Analysis, 95-107 (1994); Gotlieb et al., “Automatic Test Data Generation Using Constraint Solving Techniques,” Proc. Int'l Symposium on Software Testing and Analysis, 53-62 (1998); Gupta et al., “Automated Test Data Generation Using an Iterative Relaxation Method,” FSE 98: Foundations of Software Engineering (1998). Path coverage is notoriously difficult to work with as a coverage metric because there are an unbounded number of paths in programs with loops, which characterizes most existing non-trivial programs.
A classic problem in path-based symbolic execution is the selection of program paths. One way to guide the search for feasible paths is to execute the program symbolically along all paths, while guiding the exploration to achieve high code coverage. Clearly, it is not possible to symbolically execute all paths, so the search must be cut off at some point. Often, tools will simply analyze loops through one or two iterations. See Bush et al., “A Static Analyzer for Finding Dynamic Programming Errors,” Software—Practice and Experience, 30(7):775-802 (2000). Another way to limit the search is to bound the size of the input domain (say, to consider arrays of at most length three) or to bound the maximum path length that will be considered, as done in bounded model checking. See, e.g., Jackson et al., “Finding Bugs with a Constraint Solver,” Proc. Int'l Symposium on Software Testing and Analysis, 14-25 (2000). An experiment by Yates and Malevris provided evidence that the likelihood that a path is feasible decreases as the number of predicates in the path increases. This led them to use shortest-path algorithms to find a set of paths that covers all branches in a function. See Yates et al., “Reducing the Effects of Infeasible Paths in Branch Testing,” Proc. Symposium on Software Testing, Analysis, and Verification, 48-54 (1989).
Automatically created Boolean program abstraction can be used to analyze program behavior. See, e.g., Graf et al., “Construction of Abstract State Graphs with PVS,” CAV 97: Computer-aided Verification, 72-83 (1997); Ball et al., “Automatic Predicate Abstraction of C programs,” PLDI 01: Programming Language Design and Implementation, 203-213 (2001).
Other approaches to test generation rely on dynamic schemes. Given an existing test t, Korel's “goal-oriented” approach seeks to perturb t to a test t′ covering a particular statement, branch or path using function minimization techniques. See Korel, “Dynamic Method of Software Test Data Generation,” Software Testing, Verification and Reliability, 2(4):203-213, (1992). The potential benefit of Korel's approach is that it is dynamic and has an accurate view of memory and flow dependences. The downside of Korel's approach is that test t may be very far away from a suitable test t′.
Another approach to test generation is found in the Korat tool. See Boyapati et al., “Korat: Automated Testing Based on Java Predicates,” Proc. Int'l Symposium on Software Testing and Analysis, 123-133 (2002). This tool uses a function's precondition on its input to automatically generate all (non-isomorphic) test cases up to a given small size. It exhaustively explores the input space of the precondition and prunes large portions of the search space by monitoring the execution of the precondition. For a program that has no constraints on its input, the Korat method may not work very well. Furthermore, it requires the user to supply a bound on the input size.
Work on three-valued model checking by Bruns, Godefroid, Huth and Jagadeesan shows how to model incomplete (abstract) systems using modal transition systems (equivalently, partial Kripke Structures). See, e.g., Bruns et al., “Model Checking Partial State Spaces with 3-valued Temporal Logics,” CAV 99: Computer Aided Verification, 274-287 (1999); Godefroid et al., “Abstraction-based Model Checking Using Modal Transition Systems,” CONCUR 01: Conf. on Concurrency Theory, 426-440 (2001). Their work gives algorithms for model checking temporal logic formulae with respect to such systems.
Random or fuzz testing is another popular technique for unit testing. See, e.g., Ntafos, “On Random and Partition Testing,” Proc. Int'l Symposium on Software Testing and Analysis, 42-48 (1998).
Of course, if a designer provides a specification of the expected behavior of a software system, this specification can be used to drive test generation as well.
Although it is useful to know whether a statement in a program functions correctly when it is tested at a local level, it is also useful to know whether the behavior of the statement is still appropriate as the state of the computer program changes over time. However, because path coverage is not achievable in most programs, there exists a need for test generation techniques and tools that improve upon statement coverage or multiple condition coverage while working within a finite set of program states.
In summary, techniques and tools for achieving improved test coverage in a finite program state space are described.
In one aspect, a method of computer program test generation includes selecting a set of predicates in the computer program, calculating a set of possible predicate values for plural statements in the computer program, calculating a subset of the set of possible predicate values (e.g., possible Boolean values) for the plural statements in the computer program, and generating a test for the computer program based at least in part on the subset. The subset comprises an approximation (e.g., an under-approximation) of reachable observable states in the computer program. A Boolean program can be generated based on the selected set of predicates. A superset of the set of possible predicate values also can be calculated; the superset comprises an over-approximation of the reachable observable states in the program.
In another aspect, a set of expressions that each evaluate to a Boolean value is selected. A set of possible values for plural statements in the computer program is calculated based on the selected set of expressions. A subset of the set of possible values for the plural statements in the program is calculated, and a test for the computer program is generated using the subset. The subset comprises an approximation of possible program states for the computer program.
In another aspect, a Boolean abstraction of a program is generated. Reachability analysis of the computer program is performed, which includes finding a set of reachable observable states of the computer program. The reachability analysis is based at least in part on the Boolean abstraction. Symbolic execution is performed to generate test data for testing the computer program. The reachability analysis can include, for example, computing a lower bound of reachable observable states of the computer program, and/or computing an upper bound of the reachable observable states of the computer program.
In another aspect, an upper bound and a lower bound for a set of reachable observable states in the computer program are defined. The upper bound and the lower bound are used to form a behavior model of the computer program. Defining the lower bound can include computing a pessimistic lower bound and/or an optimistic lower bound. Test generation can be performed based at least in part on the behavior model. The test generation can include generating tests for a subset of the reachable observable states defined by the lower bound prior to generating tests for other reachable observable states. Subsequent refinement of the upper and/or lower bounds can be performed, such as by increasing the lower bound (e.g., by refining Boolean abstraction, such as by introducing additional predicates), or by decreasing the upper bound. The increased lower bound and/or decreased upper bound can then be used in test generation.
In another aspect, an over-approximated superset of reachable observable states in the computer program is determined, and an under-approximated subset of reachable observable states in the computer program is determined. The superset and the subset are used in test generation for the computer program. The superset includes each actually reachable observable state in the computer program, and the subset includes provably reachable observable states in the computer program.
In another aspect, a computer system for performing test generation for computer programs includes means for selecting a set of predicates in a computer program, means for calculating a set of possible predicate values for plural statements in the computer program, means for calculating a subset of the set of possible predicate values for the plural statements in the computer program, the subset comprising an approximation of reachable states in the computer program, and means for generating a test for the computer program using the subset.
The various techniques and tools can be used in combination or independently. Different embodiments implement one or more of the described techniques and tools.
Additional features and advantages will be made apparent from the following detailed description of different embodiments that proceeds with reference to the accompanying drawings.
Described implementations are directed to computer program test coverage improvement. Described implementations can achieve higher test coverage in computer program code testing without sacrificing efficiency.
For example, in one technique, test generation includes selecting a set of predicates in the computer program, calculating a set of possible predicate values for plural statements in the computer program, calculating a subset of the set of possible predicate values for the plural statements in the computer program, and generating a test for the computer program using the subset. The subset comprises an approximation (e.g., an under-approximation or lower bound) of reachable states in the computer program.
Although operations for the various techniques are described in a particular, sequential order for the sake of presentation, it should be understood that this manner of description encompasses minor rearrangements in the order of operations, unless a particular ordering is required. For example, operations described sequentially may in some cases be rearranged or performed concurrently. Moreover, for the sake of simplicity, flowcharts may not show the various ways in which particular techniques can be used in conjunction with other techniques. As another example, although some implementations are described with reference to specific testing tools, other tools also can be used.
The various techniques and tools can be used in combination or independently. Different embodiments implement one or more of the described techniques and tools.
I. Computing Environment
The techniques and tools described herein can be implemented on any of a variety of computing devices and environments, including computers of various form factors (personal, workstation, server, handheld, laptop, tablet, or other mobile), distributed computing networks, and Web services, as a few general examples. The techniques and tools can be implemented in hardware circuitry, as well as in software executing within a computer or other computing environment, such as shown in
With reference to
A computing environment may have additional features. For example, the computing environment 100 includes storage 140, one or more input devices 150, one or more output devices 160, and one or more communication connections 170. An interconnection mechanism (not shown) such as a bus, controller, or network interconnects the components of the computing environment 100. Typically, operating system software (not shown) provides an operating environment for other software executing in the computing environment 100, and coordinates activities of the components of the computing environment 100.
The storage 140 may be removable or non-removable, and includes magnetic disks, magnetic tapes or cassettes, CD-ROMs, CD-RWs, DVDs, or any other medium which can be used to store information and which can be accessed within the computing environment 100. For example, the storage 140 stores instructions for implementing software 180.
The input device(s) 150 may be a touch input device such as a keyboard, mouse, pen, or trackball, a voice input device, a scanning device, or another device that provides input to the computing environment 100. The output device(s) 160 may be a display, printer, speaker, CD-writer, or another device that provides output from the computing environment 100.
The communication connection(s) 170 enable communication over a communication medium to another computing entity. The communication medium conveys information such as computer-executable instructions, audio/video or other media information, or other data in a modulated data signal. By way of example, and not limitation, communication media include wired or wireless techniques implemented with an electrical, optical, RF, infrared, acoustic, or other carrier.
Techniques and tools described herein can be described in the general context of computer-readable media. Computer-readable media are any available media that can be accessed within a computing environment. By way of example, and not limitation, with the computing environment 100, computer-readable media include memory 120, storage 140, and combinations of any of the above.
Some techniques and tools herein can be described in the general context of computer-executable instructions, such as those included in program modules, being executed in a computing environment on a target real or virtual processor. Generally, program modules include functions, programs, libraries, objects, classes, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or split between program modules as desired. Computer-executable instructions may be executed within a local or distributed computing environment.
II. Techniques and Tools for Predicate-Based Testing
Many errors that go undetected in conventional program testing are due to complex correlations between program predicates (which control the execution of statements) and program statements (which in turn affect the values of the predicates). A predicate is an expression that evaluates to a Boolean value. In other words, a predicate maps a state to a Boolean value. For example, the predicate (x>0) indicates whether or not variable x has a positive value in a given state. Predicates can be drawn from the conditional statements and assertions in a program, as well as from implicit run-time “safety” checks (e.g., checking for array bound violations and division by zero) or from automated analysis, or from the programmer. Predicates can be used, for example, to generate a Boolean program abstracted from a program under test. The program under test can be a program written in C or some other programming language).
Described implementations include techniques and tools for test generation based on reachable observable states. An observable state of a computer program is an evaluation of n predicates under some program state at a statement in the program. While the set of possible program states in most programs (e.g., programs with loops) is unbounded, in a program with m statements and n predicates, the size of the set of observable states S is at most m(2^{n}). A reachable observable state is an observable state that the program being tested can actually be in. An example of an observable state that is not reachable is an observable state with an inherent logical contradiction (e.g., (x>0&& x<0)).
For example, a test coverage scenario is described that includes the concepts of over-approximation (upper bound) of reachable observable states and under-approximation (lower bound) of reachable observable states.
In described implementations, the goal of testing coverage is to minimize the difference between U and L (e.g., by increasing the lower bound), such that the observable states that must be reachable in the program represent a greater percentage of the observable states that may be reachable in the program. As the ratio |L|/|U| approaches 1.0, the testing system approaches complete coverage of reachable observable states in the program. In some embodiments, a testing system covers states in L first (i.e., prior to covering states outside of L).
III. A First Combined Implementation of Described Techniques and Tools
The following example describes techniques for using predicate abstraction and model checking to guide test case generation via symbolic execution. The features and limitations described in this example vary in other implementations.
In this example, an automated behavioral approach is defined to unit test generation of C code based on three steps: (1) predicate abstraction of the C code generates a Boolean abstraction based on a set of observations (predicates); (2) reachability analysis of the Boolean abstraction computes an over-approximation to the set of observable states and generates a small set of paths to cover these states; (3) symbolic execution of the C code generates tests to cover the paths. This approach improves upon a variety of test generation approaches based on covering code and deals naturally with the difficult issue of infeasible program paths that plagues many code-based test generation strategies.
A. Introduction
In this example, unit test generation is re-oriented to focus on covering important aspects of a program's behavior rather than its structure, and an automated process to achieve good behavioral coverage is defined. Good code coverage can be achieved as a side-effect of achieving good behavioral coverage.
The reachable states of a program are a window into its behavior. While a program may have an unbounded number of reachable states, at any point in time there usually are only a finite number of observations about these states that will be interesting to test. Predicates are a way to observe program behavior and bound the number of observed states, as there are a finite number of control locations in a program and a finite number of observations (2^{n}) that can be made at each such location. An observer o is a vector of n predicates. The observed state o(s) corresponding to state s is a vector of Boolean values (paired with a program location) constructed by applying each predicate in o to state s.
In this example, an automatic process for creating test data to achieve high behavioral coverage of a program P is defined. The input to the process is a set of predicates E observing states of P. Given these predicates, the process has three main steps, which are based on existing algorithms and tools in this example:
In this example, the above three steps are completely automatic. Selection of the predicates E is important. If E is the empty set then the Boolean abstraction will be imprecise and very few paths chosen by the model checker will be feasible. Let F(S_{E}) be the subset of paths in P(S_{E}) that are feasible paths in the source program P. The ratio |F(S_{E})|/|P(S_{E})| gives us a measure of the adequacy of a set of observations E. The closer the ratio is to one, the closer the Boolean abstraction BP(P,E) approximates the program P. Choosing the set of predicates from the conditionals in a program is a good start to achieving a high coverage ratio.
The following case study brings these ideas into focus.
B. Case Study
1. Observations
There are thirteen statements/labels in the partition function 600, but an unbounded number of paths. If the number of iterations of each loop in the function are bounded to be no greater than k, than the total number of paths is ƒ(k)=1+(2(1+k)^{2})^{k}, which grows very quickly. Which of these many paths should be tested? Which are feasible and which are infeasible? Clearly, paths are not a very good way to approach testing of this function.
Instead of reasoning in terms of paths, this example uses predicates to observe the states of the partition function 600. Let us observe the four predicates that appear in the body of the function: (lo<hi), (lo<=hi), (a[lo]<=pivot), and (a[hi]>pivot). An observed state thus is a bit vector of length four (It; le; al; ah), where lt corresponds to (lo<hi), le corresponds to (lo<=hi), al corresponds to (a[lo]<=pivot), and ah corresponds to (a[hi]>pivot). There only are ten feasible valuations for this vector, as six are infeasible because of correlations between the predicates:
Since there are thirteen labels in the code and ten possible valuations, there is a state space of 130 observable states in the worst-case. However, as explained below, the number of reachable observable states in this example is far less.
2. Boolean Abstraction
Statements in the Boolean program 700 conservatively update each Boolean variable to track the value of its corresponding predicate. The enforce statement in the Boolean program 700 has the effect of putting an assume statement (with the same expression as the enforce) before and after each statement. The expression in the enforce statement rules out the six infeasible states listed in the previous section.
Boolean programs contain parallel assignment statements. The first such assignment in the Boolean program 700 captures the effect of the statements before label L0 in the C program 600 (
lt,le,al,ah:=T,T,*,*;
This assignment statement sets the values of variables lt and le to true because the C code before label L0 establishes the conditions n>2, lo==1, and hi==n−1, which implies that lo<hi. The variables al and ah are non-deterministically assigned true or false (*) since the initial values in the input array are unconstrained.
The while loop at label L0 constrains le to be true if control passes into the body of the loop, as le is the variable corresponding to the predicate (lo<=hi). The statement lo++; at label L3 (
lt,le,al:=ch(F,!lt),ch(lt,!lt∥!le),*;
The ch function is a built-in function of Boolean programs that returns true (T) if its first argument is true, false (F) if its first argument is false and second argument is true, and * (T or F) otherwise. This translation of lo++ shows that:
The assignment statement hi−−; at label L6 (
3. Reachable States in the Boolean Abstraction
In this example, the model checker Bebop is used to compute the reachable states of the Boolean program 700. The reachable states of Boolean program 700 are shown in the table in
There are 49 reachable states in the Boolean program 700, denoted by the “x” marks in the table. This is much smaller than the total number of states, which is 130. Such sparseness is important because it rules out many states that shouldn't be covered with tests (because they are unreachable in the Boolean program 700, and thus in the C program 600 (
Consider the first four columns of the table. In each of these columns, the variable al (third bit position from left) is true. If al is true upon entry to the while loop at labels L2 to L4, then this loop iterates until al becomes false. This is why (with two exceptions) there is no state in which al is true after label L4 in the function. The exception is due to the swap procedure, which makes al true at labels LA and LB.
Now consider the next three columns (labeled TTFT, FTFT, and FFFT). In these states, the variable al is false and ah (fourth bit position from left) is true. In these states, the first inner while loop does not iterate and the second inner while loop (labels L5 to L7) will iterate until ah is false.
Finally, at the columns labeled TTFF, FFFF, and FFTT, due to the effect of the two inner while loops, the label L8 can only be reached in one of two states (TTFF or FFFF), as can be seen in the table in
In summary, the reachable state space is sparse because of correlations between predicates in the code. This sparseness makes symbolic model checking efficient. Additionally, symbolic model checking of Boolean programs has a number of advantages over directly symbolically executing a C program: (1) it can compute loop invariants over the observed predicates; (2) it is more efficient since it only observes certain aspects of the program's state. The next section explains how the state space of the Boolean program 700 can be used to effectively guide symbolic execution of the C code 600 to generate test data.
4. Feasibility Testing and Input Generation
In testing, it is desirable to generate test inputs that will cause each observed reachable state (e.g., the reachable observable states indicated in
Let us consider one of these paths: L0:TTTF→L1:TTTF→L2:TTTF→L3:TTTF→L4:TTFF→L2:TTFF and the program 1000 (
In this example, CBMC, a bounded-model checker for C programs, is used to determine whether or not the assert statement in the generated program 1000 can fail. See Clarke et al, “Behavioral Consistency of C and Verilog Programs Using Bounded Model Checking,” Design Automation Conference, 368-371 (2003).
If it can fail then it means that all the observed states in the path are reachable in the original C program 600 (
For the generated program 1000 of
At this point the increment of lo takes place and the values of the five locations now are (lo=2, hi=3, pivot=1, a[lo]=3, a[hi]=0). Control passes to label L4. The expression
((lo<hi)&&(lo<=hi)&&!(a[lo]<=pivot)&&!(a[hi]>pivot))
evaluates to true and control passes back to L2. Thus, state L2:TTFF has been covered as well.
The infeasible path is due to the fact that the state L6:FFFT is not reachable in the C program 600, but is reachable in the Boolean program 700. The path is infeasible in the C program 600 because the hi variable has been decremented so that its value is less than that of the lo variable. Because the C code 600 maintains the invariant that all array elements with index less than lo have value less than or equal to pivot, the value of a[hi] must be less than or equal to pivot. However, the state L6:FFFT requires that a[hi]>pivot is true at the end of the path. The reason the path is feasible in the Boolean program 700 is that our four chosen predicates do not track the values of the array elements that are below the index lo or are above the index hi.
As shown in
5. Analysis of the Corrected Partition Function
The partition function 600 is modified to eliminate the two array bounds violations (first and second inner while loops) and the process is re-run on the modified function. The end result of this analysis is shown in
C. Discussion
1. Predicate Selection
Our approach to test generation is parameterized by the set of predicates E, which define the precision of the Boolean abstraction BP(P,E). The more precise this abstraction, the more likely that the paths generated by the Bebop model checker (P(S_{E})) will be feasible paths (F(S_{E})) in the source program P. This gives rise to the ratio |F(S_{E})|/|P(S_{E})|, which measures the precision of the Boolean program abstraction by its ability to find feasible paths in the source program.
To illustrate this concept, consider using our test generation process on the partition function 600 (
What exactly does a |F(S_{E})|/|P(S_{E})| ratio of 1.0 signify? It says that no more tests are needed in order to cover the observable states in S_{E}. However, it doesn't mean that all interesting observations about a piece of code have been made. In our running example, the specification for the partition function 600 illustrates other observations that can be made. This specification states that at label L1 the following loop invariant holds:
for all i, 0<=i<lo, a[i]<=pivot and for all j, hi<j<n, pivot<a[j]
Generating test data based solely on the code is generally not sufficient—one should also consider the specification of what the code is supposed to do.
If a path in p_{s }generated by the model checker is infeasible in the C program, it may be that the state s at the end of p_{s }is unreachable in the C program or it may be that there is a feasible path to s but the predicates in E were insufficient to guide the model checker to this path. In this case, there are several alternatives: e.g., involve the programmer in the process to either add predicates to E to guide the model checker to a feasible path to s or assert that the state is not reachable; or use an automated tool such as SLAM to try to prove that s is reachable/unreachable in the C program, thereby generating more predicates.
2. The Small Scope Hypothesis
The “small scope hypothesis” of testing is that a high percentage of the bugs in a system can be found by exhaustively checking the program on inputs of a small size. Given this hypothesis, a central question is “how large should ‘small’ be?” Techniques in this example help to generate such small inputs that cover a set of observed states. In this example, the size of the input arrays range from length three to six. The observations E place constraints on the size of the input needed to cover the observable states S_{E}. In effect, this process can help determine how large “small” should be.
3. Example Implementation
This approach can be implemented, for example, for Microsoft's intermediate language (MSIL)—a bytecode representation now targeted by Microsoft compilers for C# and Visual Basic. A unit testing tool can be created that will automate test generation for MSIL or for some other language.
IV. A Second Combined Implementation of Described Techniques and Tools
The following example describes a new form of control-flow coverage that is based on observing the vector consisting of a program's conditional predicates, thus creating a finite-state space. The features and limitations described in this example vary in other implementations.
Consider a program with m statements and n predicates, where the predicates are derived from the conditional statements and assertions in a program, as well as from implicit run-time safety checks. An observable state is an evaluation of the n predicates under some state at a program statement.
Predicate-complete testing (PCT) seeks to cover every reachable observable state (at most (m×2^{n}) of them) in a program. PCT coverage is a new form of coverage motivated by the observation that certain errors in a program only can be exposed by considering the complex dependences between the predicates in a program and the statements whose execution they control. PCT coverage subsumes many existing control-flow coverage criteria and is incomparable to path coverage.
To support the generation of tests to achieve high PCT coverage, an upper bound U and lower bound L to the (unknown) set of reachable observable states R is defined. These bounds are constructed automatically using Boolean (predicate) abstraction over modal transition systems and can be used to guide test generation via symbolic execution. A static coverage metric is defined as |L|/|U|, which measures the ability of the Boolean abstraction to achieve high PCT coverage. Finally, a technique for increasing this ratio by the addition of new predicates is described.
A. Introduction
Control-flow-based test generation generally seeks to cover all the statements or branches in a program. There are various control-flow adequacy criteria that go beyond branch coverage, such as multiple condition coverage and path coverage. Paths are notoriously difficult to work with as a coverage metric because there are an unbounded number of them in programs with loops, which characterizes most interesting programs in existence. However, there is a growing realization that many errors that go undetected in the face of 100% statement or branch coverage are due to complex correlations between the predicates (that control the execution of statements) and the statements (that affect the value of these predicates) of a program.
An alternative to path coverage that induces a finite (rather than infinite) state space is therefore desirable. A fixed notation for atomic predicates (not containing Boolean connectives) is taken from the relevant programming language. Consider a program with m statements and n predicates. An observable state is an evaluation of the n predicates under some program state at a statement. While the set of states in a program is unbounded, the size of the set of observable states (S) is at most (m×2^{n}).
PCT testing seeks to cover all reachable observable states. The n predicates represent all the case-splits on the input that the programmer identified. (Of course, the programmer may have missed certain cases—specification-based testing can be used to determine the absence of such case-splits.) In the limit, each of the m statements may have different behavior in each of the _{2} ^{n }possible observable states, and so should be tested in each of these states. PCT coverage subsumes traditional coverage metrics such as statement, branch and multiple condition coverage and is incomparable to path coverage. PCT groups paths ending at a statement s into equivalence classes based on the observable states they induce at s.
Control-flow coverage metrics result from dividing a dynamic measure (for example, the number of statements executed by a test) into a static measure (for example, the number of statements in a program). Clearly, such a metric also can be defined for observable states. However, the choice of (m×2^{n}) as a denominator will not do, since many of the (m×2^{n}) states may be unreachable. (Statement coverage does not suffer greatly from this problem because most statements are reachable). For example, if the set of predicates contains (x=0) and (x=1) then not all combinations are possible. Furthermore, invariants established by the program will further cut down the number of reachable observable states.
Thus, to define a better denominator for PCT coverage, a technique using over-approximation and under-approximation of the set of reachable observable states (R) using the theory of modal transition systems and Boolean abstraction is described. The Boolean abstraction of a program with respect to its n predicates is a non-deterministic program, whereas the original concrete program is deterministic. In this exemplary implementation, reachability analysis of this abstract program yields an upper bound U for R (R⊂U) as well as a lower bound L for R (L⊂R). The set U is an over-approximation of R: any state outside U is not a reachable observable state and need not (indeed, cannot) be tested. This set U provides a better denominator than (m×2^{n}). Conversely, the set L is an under-approximation of R: any state in L must be a reachable observable state. States in L must be testable.
The reachability status of states in U-L is unknown. If a set of tests doesn't cover some states in L, one should first try to cover these states. L can be used to guide symbolic path-based test generation to cover the untested states in L. After covering L, the (static) ratio |L|/|U| can be brought closer to one by refining the Boolean abstraction through the introduction of additional predicates.
B. A Characterization of Predicate-complete Test Coverage
This section compares PCT coverage with other forms of control-flow coverage. In this comparison, complex predicates are decomposed into atomic predicates. So, the program fragment
L1: if ((x<0)∥(y<0)) S else T
contains two branches corresponding to the atomic predicates (x<0) and (y<0). Based on this decomposition, the concepts of branches, atomic predicates and conditions are equivalent.
Complete PCT coverage means that each reachable observable state of a program is covered by a test. This implies that each (executable) statement is executed at least once, so PCT subsumes statement coverage. PCT coverage requires that each predicate be tested so as to evaluate to both true and false (of course this may not be possible for unsatisfiable predicates such as (x!=x)), so it subsumes branch coverage. PCT clearly also subsumes multiple condition coverage and its variants. Considering the program fragment given above, multiple condition coverage requires every possible Boolean combination of (x<0) and (y<0) to be tested at L1, which seems similar to PCT. But now, consider the sequencing of two if statements:
L2: if (A∥B) S else T
L3: if (C∥D) U else V
PCT requires that every Boolean combination over the set {A, B, C, D} be tested at every statement in the program (six in this case, the two if statements and the four statements S, T, U and V). Multiple condition coverage only requires that every Boolean combination over {A, B} be tested at L2 and that every that every Boolean combination over {C, D} be tested at L3. Similarly, predicate-based test generation focuses on testing predicates in a program. See, e.g., Tai, “Theory of Fault-based Predicate Testing for Computer Programs, IEEE Trans. on Software Engineering, 22(8):552-562 (1996); Tai, “Predicate-based Test Generation for Computer Programs,” ICSE 03: Int'l Conf on Software Engineering, 267-276 (2003). It considers correlations between predicates that appear in a single conditional statement but does not consider correlations between predicates that appear in different conditional statements, as does PCT.
Of course, paths can be viewed as possible logical combinations of predicates, so it is natural to ask how PCT relates to path coverage. Since a program with n predicates can have at most 2^{n }paths, it seems like PCT might have the ability to explore more behaviors (as it may explore (m×2^{n}) states in the limit). In fact, PCT and path coverage are incomparable, even for loop-free programs.
The code listing 1300 in
{(L1, x<0), (L2, x<0), (L4, x<0), (L5,!(x<0))}
via the path (L1,L2,L4,L5), while the test {x→1} covers the observable states
{(L1,!(x<0)), (L3,!(x<0)), (L4, x<0), (L5, x<0), (L6, x<0)}
via the path (L1,L3,L4,L5,L6). However, this set of tests does not cover the feasible path (L1,L2,L4,L5,L6), which is covered by the test {x→−2}.
Because of the assignment statement “x=−2;”, the set of reachable observable states at label L4 (namely (L4,x<0)) cannot distinguish whether the executed path to L4 traversed the then or else branch of the initial if statement. While PCT can track many correlations, assignment statements such as the one above can cause PCT to lose track of correlations captured by path coverage.
In this example, if the predicate (x==−2) is added to the set of observed predicates then PCT coverage is equivalent to path coverage, as PCT coverage will require the test {x→−2 } in order to cover the reachable state (L2,x==−2).
The following code sample shows it is possible to cover all feasible paths in a (loop-free) program without covering all reachable observable states:
L1: if(p)
L2: if (q)
L3: x=0;
L4: A;
The program has three feasible paths: (L1,L2,L3,L4), (L1,L2,L4) and (L1,L4). However, a test set of size three that covers these paths clearly will miss either the observable state (L4,!p&&q) or (L4,!p&&!q).
In summary, PCT coverage is a new type of coverage criteria that subsumes statement, branch, multiple condition and predicate coverage. PCT has similarities to path coverage but is strictly incomparable, as the above examples demonstrate.
C Analysis of an Example Function
This section demonstrates upper and lower bounds to the reachable observable states of the partition function 600 (
There are thirteen labels in the partition function 600 (L0-LC), but an unbounded number of paths. Instead of reasoning in terms of paths, predicates are used to observe the states of the partition function 600. Let us observe the four predicates that appear in the conditional guards of the function: (lo<hi), (lo<=hi), (a[lo]<=pivot), and (a[hi]>pivot). An observed state thus is a bit vector of length four (lt; le; al; ah), where lt corresponds to (lo<hi), le corresponds to (lo<=hi), al corresponds to (a[lo]<=pivot), and ah corresponds to (a[hi]>pivot).
There only are ten feasible valuations for this vector, as six are infeasible because of correlations between the predicates:
1. Boolean Abstraction
As explained in Section III, the assignment statement lt,le,al,ah:=T,T,*,*; sets the values of variables lt and le to true because the C code before label L0 establishes the conditions n>2, lo==1, and hi==n−1, which implies that lo<hi. The variables al and ah are non-deterministically assigned true or false (*) since the initial values in the input array are unconstrained.
The while loop at label L0 constrains le to be true if control passes into the body of the loop, as le is the variable corresponding to the predicate (lo<=hi). The statement lo++; at label L3 translates to the parallel assignment statement in the Boolean program 1400:
lt,le,al:=(!lt?F:*), lt,*;
This translation of lo++ shows that:
The assignment statement hi−−; at label L6 (
The Boolean program 1400 shown in
2. Upper and Lower Bounds
Consider the initial state L0:TTTT. This abstract state corresponds to all concrete states that satisfy the expression:
(lo<hi)&&(a[lo]<=pivot)&&(a[hi]>pivot)
States satisfying this condition will cause the body of the outer while loop (label L1) and the body of the first inner while loop (label L3) to execute. This is reflected in the state space by the sequence of must-transitions L0:TTTT→L1:TTTT→L2:TTTT→L3:TTTT.
The set of nodes in
In summary, the ratio |L|/|U| measures the ability of the abstraction to guide test generation to cover the observable states of a program. In this example, the ratio is 43/49. If the ratio were 1.0, then we have precisely characterized the set of reachable observable states. Abstraction refinement can be used to increase the ratio |L|/|U|. As described below, it is possible to reach a ratio of 1.0 for the partition function 600 through the addition of three predicates.
D. Formalizing Abstraction
In this section, the concepts of concrete and abstract transition systems are defined that we will be used to compute the upper and lower bounds, U and L, to the set of reachable observable states R of a program.
1. Concrete Transition Systems
A deterministic sequential program is represented by a concrete transition system (CTS) as follows:
Definition 4.1: (Concrete Transition System). A concrete transition system is a triple (S_{C}; I_{C};→) where S_{C }and I_{C }are non-empty sets of states and →⊂ S_{C}×S_{C }is a transition relation satisfying the following constraints:
There are two distinguished end states, halt and error, which correspond to execution terminating normally and going wrong, respectively. These two states have no successor states. All other states have exactly one successor. Thus, a CTS models a program as a set of traces.
2. Abstract Transition Systems
Modal Transition Systems (MTSs) are a formalism for reasoning about partially defined systems that can be used to model (Boolean) abstractions of CTSs. See, e.g. Godefroid et al., “On the Expressiveness of 3-valued Models,” VMCAI 03: Verification, Model Checking and Abstract Interpretation, 206-222 (2003); Godefroid, “Reasoning About Abstract Open Systems with Generalized Module Checking,” EMSOFT 03: Conf on Embedded Software, 223-240 (2003). Modal transition systems can be generalized to tri-modal transition systems (TTSs) as follows:
Definition 4.2: (Tri-Modal Transition System). A TTS is a tuple
where S is a nonempty set of states and
are transition relations such that
A “total-onto” relation over D×E contains at least one pair (d, e), e ∈ E, for each element d ∈ D (it is total) and at least one pair (d′, e′), d′ ∈ D, for each element e′ ∈ E′ (it is onto). A total-onto abstraction relation induces an abstract TTS M_{A }from a CTS M_{C }as follows:
Definition 4.3: (Precise Abstraction Construction). Let M_{C}=(S_{C}, I_{C},→) be a CTS. Let S_{A }be a set of abstract states and ρ be a total-onto abstraction relation over pairs of states in S_{C}×S_{A}. Let match+ and match− relate states in S_{C}×S_{A }as follows:
match+(S _{c} , S′ _{a})=∃(S′ _{c} , S′ _{a})∈ρ:S _{c} →S′ _{c }
match−(S′ _{c} , S _{a})=∃(S _{c} , S _{a})∈ρ:S _{c} →S′ _{c }
is constructed from M_{C}, S_{A }and ρ as follows:
It is easy to see that the definition of M_{A }satisfies the constraints of a TTS, namely that
The if-and-only-if (“iff”) notation indicates the assumption that it is possible to create a most precise abstract TTS M_{A }from a given CTS M_{C}. In general, this assumption does not hold for infinite-state M_{C}. It does hold for the partition function and other code examples we consider here.
In
transitions (solid line edges) and may-transitions (dotted edges) are present.
3. Predicate Abstraction
Predicate abstraction maps a (potentially infinite-state) CTS into a finite-state TTS defined by a finite set of quantifier-free formulas of first-order logic Φ={φ_{l}, . . . , φ_{n}}. A bit vector b of length n (b=b_{l }. . . b_{n}, b_{i}∈{0, 1}) defines an abstract state whose corresponding concrete states are those satisfying the conjunction (b,Φ)=(l_{l}^ . . . ^ l_{n}) where l_{i}=φ_{i }if b_{i}=1 and l_{i}=
φ_{n}if b_{i}=0. s|=(b,Φ) denotes that (b, φ) holds in state s.Definition 4.4: (Predicate Abstraction of a CTS). Given a CTS M_{C}=(S_{C}, I_{C}, →) and a set of predicates Φ={φ_{l}, . . . , φ_{n}}, predicate abstraction defines the total-onto abstraction relation ρ and the set of abstract states S_{A}:
(per Definition 4.3). Assume that S_{A }contains abstract states halt_{A }and error_{A }that are in a one-to-one relationship with their counterparts halt and error from S_{C}.
Algorithms for computing the may and must+ transitions of a predicate abstraction of an MTS are given by Godefroid, Huth and Jagadeesan. See Godefroid et al., “Abstraction-based Model Checking Using Modal Transition Systems,” CONCUR 01: Conf. on Concurrency Theory, 426-440 (2001). Computation of the must− transitions can be done in a similar fashion. Computation of the most precise abstract transitions is undecidable (neither formally provable nor unprovable), in general. Assume here the existence of a complete theorem prover that permits the computation of the most precise abstract transitions.
Let WP(s, e) be the weakest pre-condition of a statement s with respect to expression e and let SP(s, e) be the strongest post-condition of s with respect to e. See Gries, The Science of Programming, Springer-Verlag (1981). (For any state c, satisfying WP(s, e) the execution of s from c_{1 }results in a state c_{2 }satisfying e. For any state c_{1 }satisfying e the execution of s from c_{1 }results in a state c_{2 }satisfying SP(s, e)). Let P_{1 }and P_{2 }be the concretization of two bit vectors b_{1 }and b_{2 }(i.e., P_{1}=(b_{1}, Φ>) and P_{2}=<(b_{2}, Φ<). Statement s induces a may-transition from b_{1 }to b_{2 }if ∃V: P_{1}^ WP(s, P_{2}), where V is the set of free variables in the quantified expression. Statement s induces a must+ transition from b_{1 }to b_{2 }if ∀V.P_{2}
SP(s, P_{1}).In order to show relations between the reachable states of M_{A }and M_{C}, it is useful to define a concretization function γ mapping states in S_{A }to states in S_{C}:
Definition 4.5: (Concretization Function). Let ρ:S_{C}×S_{A }be an abstraction relation. Let
γ_{ρ}(A)={S _{c} |∃S _{a} ∈A:(S _{c} , S _{a})∈ρ}
be the concretization function mapping a set of abstract states to its corresponding set of concrete states.
Using this function, the set of initial abstract states I_{A }of M_{A }is defined as the set satisfying the formula
I _{C}=γ_{ρ}(I _{A})
When ρ is understood from context we will use γ rather than γ_{ρ}.
(bold edge 1710 in graph 1700). For similar reasons, we have the must# transition
(L1, !(x<0))(bold edge 1720).
The next statement in program 1300 is the if-statement at label L1. Because this statement branches exactly on the predicate (x<0), it induces the following must# transitions, indicated by bold edges 1730 and 1740, respectively:
The statement at label L2 in program 1300 is a skip and so has no affect on the state, inducing the transition
(bold edge 1750).
The assignment statement at label L3 in program 1300 is reachable only when ! (x<0) is true. It assigns the value −2 to variable x. We have that WP(x=−2, (x<0))=(−2<0), which reduces to true. This means that there is a must+ transition
(solid edge 1760 marked with a “+” symbol). However, WP(x=−2, !(x<0))=(!(−2<0)), which reduces to false. So there can be no transition from (L3, !(x<0)) to (L4, !(x<0)). Now, consider strongest post-conditions. We have that SP(x=−2, !(x<0))=!(−2<0), which reduces to false, so there can be no must− transition from (L3, !(x<0)) to (L4, (x<0).
Now consider the assignment statement at label L4 in program 1300 which is reachable only under (x<0) and which increments variable x. Because SP(x=x+1, (x<0)=(x<1) and the set of states satisfying (x<0) is a subset of the set of states satisfying (x<1), there is a must− transition
(solid edge 1770 marked with a “−” symbol). There is no must+ transition between these states because WP(x=x+1, (x<0)=(x<−1)) and the set of states satisfying (x<0) is not a subset of the set of states satisfying (x<−1). The assignment statement induces a may transition
(dotted edge 1780) because this transition only takes place when variable x has the value −1 before the increment and the (resulting) value 0 after the increment.
Finally, there is a must# transition
(bold edge 1790) because the if-statement at label L5 tests exactly the condition (x<0).
E. Defining Predicate-Complete Test Coverage
Recall that the goal of predicate-complete testing (PCT) is to cover reachable states, as defined by the m statements and n predicates Φ={φ_{1}, . . . , } in the program represented by the CTS M_{C}. The set of reachable observable states R is unknown, so the Boolean (predicate) abstraction of M_{C }will be used with respect to Φ to construct an abstract TTS M_{A }(see Definition 4.4).
In this section, M_{A }is analyzed to compute both upper and lower bounds to R. To do so, it useful to define a reachability function for a transition system. Let S be a set of states and δ be a transition relation of type S×S. The reachability function over δ and X⊂S is defined as the least fixpoint of:
reach[δ](X)=X∪reach[δ](δ(X))
where δ (X) is the image of set X under δ.
Reachability in a CTS: Let M_{C }be a CTS. The set of states reachable from states in T (T ⊂ S_{C}) is denoted as:
reach_{C}(T)=reach[→](T)
That is, reachability in M_{C }is simply defined as the transitive closure over the transitions in M_{C}, starting from states in T.
1. Upper Bound Computation
May-reachability in TTS M_{A }defines the upper bound U. Let M_{C }be a CTS and let M_{A }be an abstract TTS defined by abstraction relation ρ (via Definition 4.3). The upper bound is defined as:
That is, U is simply defined as the transitive closure over the may transitions in M_{A }from the initial states I_{A}. It is easy to see that reach_{C}(I_{C})⊂γ(U), as the may transitions of M_{A }over-approximate the set of transitions in M_{C }(by Definition 4.3). Since the reachable observable states R are contained in reach_{C}(I_{C}), we have that U is an over-approximation of R.
2. Pessimistic Lower Bound Computation L_{p }
A set of abstract states X ⊂ S_{A }is a lower bound of R ⊂ S_{C }if for each x_{a }∈ X, there is a (x_{c}; x_{a}) ∈ ρ such that x_{c }E reach_{C}(I_{C}).
The computation of the lower bound can be presented in two steps. First, the pessimistic lower bound (L_{p}) is defined, which makes no assumptions about M_{C}. Second, the optimistic lower bound (L_{o}) is defined, which assumes that M_{C }does not diverge.
L_{p }is defined as:
That is, an abstract state v_{a }is in L_{p }if there is a (possibly empty) sequence of must− transitions leading from s_{a}∈I_{A }to t_{a}, there is a may transition from t_{a }to u_{a }(or t_{a }is equal to u_{a}), and there is a (possibly empty) sequence of must+ transitions from u_{a }to v_{a}.
For each v_{a }∈ L_{p}, there is a (v_{c}, v_{a}) ∈ ρ such that v_{c }∈ reach_{C}(I_{C}). That is, L_{p }is a lower bound to R. The proof is done in three steps, corresponding to the three parts of the definition of L_{p}:
The optimistic computation of the lower bound (L_{o}) assumes that the program under consideration doesn't diverge (contain an infinite loop). The problem of detecting nontermination is left to testing (which is typically done by defining some timeout threshold after which a computation is declared to be non-terminating).
States that must be reached (assuming convergence) are captured using the idea of post-dominance. In our context, a state t_{a }post-dominates s_{a }in M_{A }if every path of may transitions from s_{a }to a vertex in the set {halt_{A}, error_{A}} contains t_{a}. Post-dominance in M_{A }is defined as the greatest fixpoint to the following set of equations:
Using L_{p }and pd_{A}, L_{o }is defined as follows:
L _{o} ={x _{a} |x _{a} ∈pd _{A}(W _{a})Λw _{a} ∈L _{p}}
It is easy to prove that if M_{C }contains no diverging computations then for all states w_{c }mapping to w_{a}, if x_{a}∈pd_{A}(W_{a}) then there is a state x_{c }mapping to x_{a }such that x_{c }∈ reach_{C}({w_{c}}). Therefore, assuming that M_{C }contains no diverging computations, for each x_{a }∈ L_{o}, there is an x_{c }mapping to x_{a }such that v_{c }∈ reach_{C}(I_{C}) The lower bound L is defined as L=L_{o}.
4. Application to Partition Function Example
Let M_{A }be the TTS constructed from concrete TTS M_{C}(via predicate abstraction on the set of predicates Φ, which induces an abstraction relation ρ). Let I_{C }be the set of initial concrete states of M_{C }and let I_{A }be the corresponding set of initial abstract states of M_{A }(induced by ρ).
Referring again to
I _{A} ={L0:TTFT, L0:TTTT, L0:TTTF, L0:TTFF}
In
is one of the paths that leads to an array bounds error in partition function 600. Note that in this path the label L3 is covered twice, once by the state L3:TTTF and then by the state L3:FTTF. In the first state, lo<=hi, a[lo]<=pivot and a[hi]<=pivot. At label L3, lo is incremented by one. The path dictates (via covering the may transition
that the value of lo and hi are now equal. Because a[hi]<=pivot the loop at label L2 continues to iterate and the second state is reached, L3:FTTF, in which lo==hi and a[lo]<=pivot and a[hi]<=pivot. When lo is incremented the second time, its value becomes greater than hi, whose value still is the index of the last element of the array. Thus, the next access of a[lo] (see state L2:FFFF) will cause an array bounds violation.
F. Test Generation
The goal of test generation is to cover all the states in the lower bound L (plus any additional states). Our test generation process consists of three steps:
1. Path Generation
Let I_{A }be the set of initial abstract states in M_{A}. Consider the set of states L_{p}. The goal of the path generation phase is to enumerate all paths from I_{A }consisting of a sequence of must− transitions followed by one (and perhaps no) may transition, while covering no state more than once. This can be done by a simple depth-first search procedure. The idea is that if tests are generated to cover these paths then the rest of the states in L_{o }will be covered if the execution of program does not go wrong (uncover an error) or diverge.
In
2. Symbolic Execution
Each of the ten paths represented in table 1800 induces a straight-line C “path” program that is automatically generated by tracing the path through the partition function 600 (
L0:TTTF→L1:TTTF→L2:TTTF→L3:TTTF→L4:TTFF
and its corresponding path program 1900 in
In this example, the CBMC bounded-model checker is used to generate a counterexample to the assertion that the state L4:TTFF cannot occur. CBMC produces an input array a[ ] and array length n that will cause the assert statement to fail, proving that L4:TTFF is reachable. For the generated path program 1900 of
3. Observe Test Runs
Instrumentation of the original program both collects the executed observable states for each test run and checks for array bounds violations. In this example, there are ten runs, two of which produce array bounds violations (because the lo index is incremented past the end of the input array and then a[lo] is accessed), as shown in the third column of
The set of observed states resulting from executing all ten tests contains all the states in
Fixing the error in the program and rerunning the entire process results in an upper bound U with 56 states and a lower bound L, of 37 states. Test generation succeeds in covering all 37 states in the lower bound L and causes no array bounds errors. Additionally, these tests cover six of the 19 tests in U-L. This leads us to consider whether or not the remaining states in U-L are reachable at all and to the problem of refining the upper and lower bounds.
G. Refinement of Lower and Upper Bounds
We now consider the problem of bringing the lower bound L and upper bound U closer together. We focus our attention on the observable states in U-L that were not covered by the test generation process of the previous section. The main question we wish to answer for these states is whether or not they are reachable in the original program. Automated machinery (such as the SLAM toolkit) can be used to try and answer this undecidable problem, but for many cases, we will need to involve the programmer or tester.
Consider the state L7:FFFT from
lo>hi && a[lo]>pivot && a[hi]>pivot
Notice that the partition function 600 (
Rather than ignoring abstract states whose concrete counterparts are unreachable, it is important to introduce new predicates to try and eliminate such states in the abstraction. The reason is that these unreachable states often will point to boundary conditions that have not yet been tested. In order to eliminate the state L7:FFFT, three new predicates can be introduced into the Boolean abstraction (in addition to the four already there) in order to track the status of the array when the variable lo takes on the value hi+1:
(lo==hi+1), (a[lo−1]<=pivot), (a[hi+1]>pivot)
These predicates track an important boundary condition that was not observed by the initial four predicates.
With these additional predicates, the generated Boolean abstraction has matching lower and upper bounds (actually L_{p}=U) and the test generation process covers all reachable observable states. As mentioned before, we cannot expect to be able to achieve matching lower and upper bounds in every case.
H. Discussion of Some Aspects of Second Combined Implementation
We now consider what it means when set of states in the upper bound U and the pessimistic lower bound L_{p }are the same (U=L_{p}). We refer to this condition as “state simulation,” as it means that every abstract state in the upper bound is observable by some execution of the concrete program. We find it useful to informally describe the states represented by these two sets using regular expressions:
That is, U is the set of abstract states reachable (from the initial set of abstract states I_{A}) via a sequence of may transitions, while L_{p }is the set of states reachable from I_{A }via a sequence of must− transitions, followed by at most one may-transition, followed by a sequence of must+ transitions.
Given a path p_{a }of abstract transitions in M_{A }containing more than one may transition (not matched by a must transition), it is impossible to know (without analysis of M_{C}) whether or not there exists a corresponding feasible execution path p_{c }in M_{C}. This is why the definition of L_{p }permits at most one may transition (not matched by a must transition).
State simulation is weaker than bi-simulation, which for deterministic systems (as we consider here) reduces to trace equivalence. See Milner, Communicating and Mobile Systems: The pi-Calculus, Cambridge University Press (1999). An abstract TTS M_{A }bi-simulates a CTS M_{C }if each may transition in M_{A }is matched by a must+ transition
It is easy to see that if M_{A }bi-simulates M_{C }then every abstract state in U is reachable via a sequence of must+ transitions. Therefore, U=L_{p}. Our use of must− transitions followed by at most one may-transition is the way in which we weaken bi-simulation to state simulation.
Under state simulation, every abstract state s_{a }in U is reachable via a sequence of must− transitions, followed by at most one may transition, followed by a sequence of must+ transitions (which characterizes L_{p}). As we have shown previously, the existence of this sequence in M_{A }implies the existence of an execution trace in M_{C }in which s_{a }is observed. This implies that there is a finite set of tests sufficient to observe all states in U. Since U is an upper bound to the set of reachable observable states R this set of tests covers all states in R as well (that is, R=U=L_{p}).
The upper bound U is, by construction, sound (that is, R⊂U). State simulation implies that the least fixpoint of the may-abstraction induced by the set of observation predicates is complete (that is, U⊂R)—or, more precisely, relatively complete, since we assume that abstract transitions can be computed precisely. In other words, the set U is equal to the set of observable states that would be encountered during the (infinite) computation of the least fixpoint over the concrete transition system M_{C }(with respect to the set of initial states I_{C}). This follows from the fact that L_{p}=U. It follows that state simulation is a sufficient test for determining the completeness of a may abstraction.
To summarize, the condition of state simulation (U=L_{p}) joins together the worlds of abstraction and testing. State simulation implies both a sound and complete abstract domain that can be completely covered by a finite set of tests.
V. Comparison of Described Techniques and Tools with Other Work
Various implementations of the techniques and tools described herein guide test generation using Boolean (predicate) abstraction and model checking, and compute upper and lower bounds to the set of reachable observable states, improving on path-based testing. Various implementations of the techniques and tools described herein use the set of input predicates to bound the set of paths that will be used to generate test data for a program. The predicates induce a Boolean abstraction that guides the selection of paths.
Various implementations of the techniques and tools described herein improve on three-valued model checking with computation of lower and upper bounds. Computation of lower and upper bounds infers properties of a concrete TTS M_{C }from analysis of an abstract TTS M_{A}. The lower bound L characterizes those observable states that are definitely reachable, the upper bound U (more precisely, its inverse S-U) characterizes those observable states that are definitely not reachable, and the reachability status of states in U-L are unknown. For example, to achieve a precise lower bound, the definition of must transitions given for MTS can be generalized to account for three types of must transitions: must+ (which correspond to must transitions in an MTS), must− and must#. In model checking of abstractions of concrete transition systems, one is interested in proving that a temporal property holds for all concrete execution paths starting from some initial abstract state. This is the reason why only must+− transitions are used in model checking of modal transitions systems. For reachability, one is interested proving the existence of some concrete execution path starting from some initial abstract state. Thus, must− transitions are of interest.
Other alternatives to path coverage include linear code sequence and jump (LCSAJ) coverage and data-flow coverage based on def-use pairs. An LCSAJ represents a sequence of statements (which may contain conditional statements) ending with a branch. An LCSAJ is an acyclic path (no edge appears twice) through a control-flow graph ending with a branch statement. PCT coverage is incomparable to path coverage for loop-free programs, so it also is incomparable to LCSAJ coverage. The goal of def-use coverage is to cover, for each definition d of a variable x and subsequent use u of variable x, a path from d to u not containing another definition of x. If there is such a path from d to u then there is an acyclic path from d to u that doesn't contain another definition of x, so again PCT coverage is incomparable to def-use coverage.
Having described and illustrated the principles of our innovations in the detailed description and the accompanying drawings, it will be recognized that the various embodiments can be modified in arrangement and detail without departing from such principles. It should be understood that the programs, processes, or methods described herein are not related or limited to any particular type of computing environment, unless indicated otherwise. Various types of general purpose or specialized computing environments may be used with or perform operations in accordance with the teachings described herein. Elements of embodiments shown in software may be implemented in hardware and vice versa.
Cited Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|
US5615359 * | Jun 23, 1994 | Mar 25, 1997 | Candle Distributed Solutions, Inc. | Data server with data probes employing predicate tests in rule statements |
US5937195 * | Nov 27, 1996 | Aug 10, 1999 | Hewlett-Packard Co | Global control flow treatment of predicated code |
US6463582 * | Oct 21, 1998 | Oct 8, 2002 | Fujitsu Limited | Dynamic optimizing object code translator for architecture emulation and dynamic optimizing object code translation method |
US6694290 * | May 25, 1999 | Feb 17, 2004 | Empirix Inc. | Analyzing an extended finite state machine system model |
US6725431 * | Jun 30, 2000 | Apr 20, 2004 | Intel Corporation | Lazy symbolic model checking |
US6834383 * | Nov 26, 2001 | Dec 21, 2004 | Microsoft Corporation | Method for binary-level branch reversal on computer architectures supporting predicated execution |
US6853963 * | May 25, 1999 | Feb 8, 2005 | Empirix Inc. | Analyzing an extended finite state machine system model |
US6898784 * | Sep 15, 2000 | May 24, 2005 | Nortel Networks Limited | Method and system for forming skeletons for generating verification systems |
US6904590 | May 25, 2001 | Jun 7, 2005 | Microsoft Corporation | Methods for enhancing program analysis |
US6944848 * | May 3, 2001 | Sep 13, 2005 | International Business Machines Corporation | Technique using persistent foci for finite state machine based software test generation |
US7055065 * | Sep 5, 2001 | May 30, 2006 | International Business Machines Corporation | Method, system, and computer program product for automated test generation for non-deterministic software using state transition rules |
US7058925 | Apr 30, 2002 | Jun 6, 2006 | Microsoft Corporation | System and method for generating a predicate abstraction of a program |
US7117484 * | Apr 16, 2002 | Oct 3, 2006 | International Business Machines Corporation | Recursive use of model based test generation for middleware validation |
US20020129343 * | Dec 28, 2000 | Sep 12, 2002 | International Business Machines Corporation | Estimation of object lifetime using static analysis |
US20030014734 * | May 3, 2001 | Jan 16, 2003 | Alan Hartman | Technique using persistent foci for finite state machine based software test generation |
US20030196191 * | Apr 16, 2002 | Oct 16, 2003 | Alan Hartman | Recursive use of model based test generation for middlevare validation |
US20030204834 * | Apr 30, 2002 | Oct 30, 2003 | Microsoft Corporation | System and method for generating a predicate abstraction of a program |
US20040226006 * | May 5, 2003 | Nov 11, 2004 | Jeffry Russell | Program slicing for codesign of embedded systems |
US20040255275 * | Jun 11, 2003 | Dec 16, 2004 | Microsoft Corporation | Method and system for generating an efficient test suite from a domain description with given constraints |
US20050076331 * | Oct 2, 2003 | Apr 7, 2005 | Microsoft Corporation | Source code annotation language |
Reference | ||
---|---|---|
1 | "Compiling to MSIL", http://msdn.microsoft.com/library/en-us/cpguide [Downloaded from the World Wide Web on Dec. 16, 2005]. | |
2 | "Halting Problem-From Math World", http://mathworld.wolfram.com/HaltingProblem.html [Downloaded from the World Wide Web on Dec. 16, 2005]. | |
3 | Allen, "Control Flow Analysis," SIGPLAN Notices, 5(7):1-19 (1970). | |
4 | Ball et al., "Automatic Predicate Abstraction of C Programs," PLDI 01: Programming Language Design and Implementation, 203-213 (2001). | |
5 | Ball et al., "Bebop: A Symbolic Model Checker for Boolean Programs," SPIN 00: SPIN Workshop, 113-130 (2000). | |
6 | Ball et al., "The SLAM Project: Debugging System Software Via Static Analysis," POPL 02: Principles of Programming Languages, 1-3 (Jan. 2002). | |
7 | Beyer et al., "Generating Tests from Counterexamples," Proc. 26th Int'l Conf. on Software Eng'g, Edinburgh, Scotland (May 23-28, 2004). | |
8 | * | Boyapati et al. "Korat: Automated Testing Based on Java Predicates," International Symposum on Software Testing and Analysis, 123-133 (2002) hereinafter Korat. |
9 | Boyapati et al., "Korat: Automated Testing Based on Java Predicates," International Symposium on Software Testing and Analysis, 123-133 (2002). | |
10 | Boyer et al., "SELECT-A Formal System for Testing and Debugging Programs by Symbolic Execution," SIGPLAN Notices, 10(6):234-245 (1975). | |
11 | Bruns et al., "Model Checking Partial State Spaces With 3-Valued Temporal Logics," CAV 99:Computer Aided Verification, 274-287 (1999). | |
12 | Bush et al., "A Static Analyzer for Finding Dynamic Programming Errors," Software-Practice and Experience, 30(7):775-802 (2000). | |
13 | Clarke et al., "Behavioral Consistency of C and Verilog Programs Using Bounded Model Checking," Design Automation Conference, 368-371 (Jun. 2003). | |
14 | Clarke, "A System to Generate Test Data and Symbolically Execute Programs," IEEE Trans. On Software Engineering, 2(3):215-222 (1976). | |
15 | Cytron et al., "Compact Representations for Control Dependence", Proceedings of ACM SIGPLAN 1990, Conference on Programming Language Design and Implementation, pp. 337-351 (1990). | |
16 | de Alfaro et al., "Three-valued Abstractions of Games: Uncertainty, but with Precision," LICS 04: Proc. 19th IEEE Symposium on Logic in Computer Science, 10 p. (Jul. 2004). | |
17 | Duran et al., "A Report on Random Testing", Proceedings of 5th International IEEE Conference on Software Engineering, 179-183 (1981). | |
18 | Giacobazzi et al., "Making Abstract Interpretations Complete," Journal of the ACM, 47(2):361-416 (2000). | |
19 | Godefroid et al., "Abstraction-based Model Checking Using Modal Transition Systems," CONCUR 01: Conference on Concurrency Theory, 426-440 (2001). | |
20 | Godefroid et al., "On the Expressiveness of 3-Valued Models," VMCAI 03: Verification, Model Checking and Abstract Interpretation, 206-222, New York (Jan. 2003). | |
21 | Godefroid, "Reasoning About Abstract Open Systems with Generalized Module Checking," EMSOFT 03: Conference on Embedded Software, 223-240, Philadelphia (Oct. 13-15, 2003). | |
22 | Goodenough et al., "Toward a Theory of Test Data Selection", IEEE Trans. Software Engineering, 1 (2):156-173 (1976). | |
23 | Gotlieb et al., "Automatic Test Data Generation Using Constraint Solving Techniques," Proceedings of International Symposium on Software Testing and Analysis, 53-62 (1998). | |
24 | Graf et al., "Construction of Abstract State Graphs With PVS," CAV 97: Computer-Aided Verification, 72-83 (1997). | |
25 | Gries, The Science of Programming, Springer-Verlag (1981). | |
26 | Gupta et al., "Automated Test Data Generation Using an Iterative Relaxation Method," FSE 98: Foundations of Software Engineering (1998). | |
27 | Harder et al., "Improving Test Suites via Operational Abstraction," ICSE 2003: Int'l Conf. on Software Eng'g, Portland, Oregon, pp. 60-71 (May 6-8, 2003). | |
28 | Howden, "Reliability of the Path Analysis Testing Strategy," IEEE Trans. on Software Engineering, 2:208-215 (1976). | |
29 | Jackson et al., "Finding Bugs with a Constraint Solver," Proc. International Symposium on Software Testing and Analysis, 14-25 (2000). | |
30 | Jasper et al., "Test Data Generation and Feasible Path Analysis," Proc. International Symposium on Software Testing and Analysis, 95-107 (1994). | |
31 | Korel, "Dynamic Method of Software Test Data Generation," Software Testing, Verification and Reliability, 2(4):203-213 (1992). | |
32 | Lee et al., "Principles and Methods of Testing Finite State Machines-A Survey", Proceedings of IEEE, 84 (8): 1090-1123 (Aug. 1996). | |
33 | Miller et al., "An Empirical Study of the Reliability of UNIX Utilities", Comm. of ACM, 33 (12):32-44 (1990). | |
34 | Milner, "Communicating and Mobile Systems: the pi Calculus", Press Syndicate, University of Cambridge, pp. | |
35 | Ntafos, "On Random and Partial Testing," Proc. International Symposium on Software Testing and Analysis, 42-48 (1998). | |
36 | Ramamoorthy, et al., "On the Automated Generation of Program Test Data", IEEE Trans. Software Engineering, 2 (4):293-300 (Dec. 1976). | |
37 | Shoham et al., "Monotonic Abstraction-Refinement for CTL," TACAS 04: Tools and Algorithms for Construction and Analysis of Systems, LNCS 2988, pp. 546-560, Barcelona (Apr. 2004). | |
38 | Tai, "Predicate-Based Test Generation for Computer Programs," Proc. 15th Int'l Conf. on Software Eng'g, 267-276 (1993). | |
39 | Tai, "Theory of Fault-Based Predicate Testing for Computer Programs", IEEE Trans. Software Engineering, 22(8):552-562 (1996). | |
40 | * | Tai, Kuo-Chung. "Predicate-Based Test Generation for Computer Programs." International conference on software engineering 15th(1993): 267-276. |
41 | Yates et al., "Reducing the Effects of Infeasible Paths in Branch Testing," Proc. Symposium on Software Testing, Analysis and Verification, 48-54 (1989). |
Citing Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|
US7624379 * | Nov 24, 2009 | Advanced Testing Technologies, Inc. | Test program set obsolescence mitigation through software and automatic test equipment system processes | |
US7908590 * | Mar 15, 2011 | Parasoft Corporation | System and method for automatically creating test cases through a remote client | |
US7921411 * | Apr 5, 2011 | International Business Machines Corporation | Model checking of non-terminating software programs | |
US8108826 * | Sep 29, 2004 | Jan 31, 2012 | Avaya Inc. | Code-coverage guided prioritized test generation |
US8271253 * | Sep 18, 2012 | International Business Machines Corporation | Symbolic depth-first searches using control flow information for improved reachability analysis | |
US8302080 * | Oct 10, 2008 | Oct 30, 2012 | Ntt Docomo, Inc. | Automated test input generation for web applications |
US8370808 * | Feb 5, 2013 | Kabushiki Kaisha Toshiba | Apparatus and a method for generating a test case | |
US8387020 * | Feb 26, 2013 | Google Inc. | Verifying that a method has been implemented for a program written in a dynamic programming language | |
US8402440 * | Feb 26, 2009 | Mar 19, 2013 | Nec Laboratories America, Inc. | Program verification through symbolic enumeration of control path programs |
US8479171 | May 24, 2010 | Jul 2, 2013 | Fujitsu Limited | Generating test sets using intelligent variable selection and test set compaction |
US8484618 | Nov 23, 2011 | Jul 9, 2013 | Advanced Testing Technologies, Inc. | Test program set obsolescence mitigation through software and automatic test equipment processes |
US8495581 | May 5, 2011 | Jul 23, 2013 | International Business Machines Corporation | Code evaluation based on failure prediction |
US8504997 * | Mar 19, 2009 | Aug 6, 2013 | Fujitsu Limited | Environment data refinement based on static analysis and symbolic execution |
US8539448 * | Sep 27, 2011 | Sep 17, 2013 | Abb Ag | System and method for automatically testing a program for safety-related automation systems |
US8549486 * | Apr 21, 2008 | Oct 1, 2013 | Microsoft Corporation | Active property checking |
US8601446 * | Oct 22, 2008 | Dec 3, 2013 | Wipro Limited | Complexity dependency change impact system and method |
US8661454 | Jan 26, 2010 | Feb 25, 2014 | Target Brands, Inc. | System and method for receiving and transmitting event information |
US8752021 * | Nov 8, 2012 | Jun 10, 2014 | Concurix Corporation | Input vector analysis for memoization estimation |
US8769500 * | Dec 1, 2010 | Jul 1, 2014 | Fujitsu Limited | Node computation initialization technique for efficient parallelization of software analysis in a distributed computing environment |
US8789054 * | Nov 30, 2010 | Jul 22, 2014 | Fujitsu Limited | Scheduling policy for efficient parallelization of software analysis in a distributed computing environment |
US8850406 * | Apr 5, 2012 | Sep 30, 2014 | Google Inc. | Detecting anomalous application access to contact information |
US8909992 | Nov 27, 2012 | Dec 9, 2014 | International Business Machines Corporation | Dynamic concolic execution of an application |
US8918772 | Jul 25, 2007 | Dec 23, 2014 | Google Inc. | Statically analyzing program correctness for a dynamic programming language |
US8984638 | Nov 12, 2013 | Mar 17, 2015 | Fireeye, Inc. | System and method for analyzing suspicious network data |
US8990939 | Jun 24, 2013 | Mar 24, 2015 | Fireeye, Inc. | Systems and methods for scheduling analysis of network content for malware |
US8990944 | Feb 23, 2013 | Mar 24, 2015 | Fireeye, Inc. | Systems and methods for automatically detecting backdoors |
US8996922 * | Nov 21, 2012 | Mar 31, 2015 | Fujitsu Limited | Mixed numeric and string constraint analysis |
US8997219 | Jan 21, 2011 | Mar 31, 2015 | Fireeye, Inc. | Systems and methods for detecting malicious PDF network content |
US9009822 | Feb 23, 2013 | Apr 14, 2015 | Fireeye, Inc. | Framework for multi-phase analysis of mobile applications |
US9009823 | Feb 23, 2013 | Apr 14, 2015 | Fireeye, Inc. | Framework for efficient security coverage of mobile software applications installed on mobile devices |
US9015674 * | Sep 28, 2012 | Apr 21, 2015 | Microsoft Technology Licensing, Llc | Identifying execution paths that satisfy reachability queries |
US9037916 | Sep 12, 2013 | May 19, 2015 | International Business Machines Corporation | Dynamic concolic execution of an application |
US9098621 | Feb 28, 2011 | Aug 4, 2015 | Microsoft Technology Licensing, Llc | Modeling software behavior using learned predicates |
US9118715 | May 10, 2012 | Aug 25, 2015 | Fireeye, Inc. | Systems and methods for detecting malicious PDF network content |
US9159035 | Feb 23, 2013 | Oct 13, 2015 | Fireeye, Inc. | Framework for computer application analysis of sensitive information tracking |
US9176843 | Feb 23, 2013 | Nov 3, 2015 | Fireeye, Inc. | Framework for efficient security coverage of mobile software applications |
US9223972 | Mar 31, 2014 | Dec 29, 2015 | Fireeye, Inc. | Dynamically remote tuning of a malware content detection system |
US9225740 | Sep 24, 2014 | Dec 29, 2015 | Fireeye, Inc. | Framework for iterative analysis of mobile software applications |
US9262635 | Feb 5, 2014 | Feb 16, 2016 | Fireeye, Inc. | Detection efficacy of virtual machine-based analysis with application specific events |
US9282109 | Jun 30, 2014 | Mar 8, 2016 | Fireeye, Inc. | System and method for analyzing packets |
US9294501 | Sep 30, 2013 | Mar 22, 2016 | Fireeye, Inc. | Fuzzy hash of behavioral results |
US9300686 | Jul 18, 2013 | Mar 29, 2016 | Fireeye, Inc. | System and method for detecting malicious links in electronic messages |
US9306960 | Aug 19, 2013 | Apr 5, 2016 | Fireeye, Inc. | Systems and methods for unauthorized activity defense |
US9306974 | Feb 11, 2015 | Apr 5, 2016 | Fireeye, Inc. | System, apparatus and method for automatically verifying exploits within suspect objects and highlighting the display information associated with the verified exploits |
US9311479 | Mar 14, 2013 | Apr 12, 2016 | Fireeye, Inc. | Correlation and consolidation of analytic data for holistic view of a malware attack |
US9355247 | Mar 13, 2013 | May 31, 2016 | Fireeye, Inc. | File extraction from memory dump for malicious content analysis |
US9356944 | Jun 28, 2013 | May 31, 2016 | Fireeye, Inc. | System and method for detecting malicious traffic using a virtual machine configured with a select software environment |
US9363280 | Aug 22, 2014 | Jun 7, 2016 | Fireeye, Inc. | System and method of detecting delivery of malware using cross-customer data |
US9367681 * | Feb 23, 2013 | Jun 14, 2016 | Fireeye, Inc. | Framework for efficient security coverage of mobile software applications using symbolic execution to reach regions of interest within an application |
US9398028 | Jun 26, 2014 | Jul 19, 2016 | Fireeye, Inc. | System, device and method for detecting a malicious attack based on communcations between remotely hosted virtual machines and malicious web servers |
US9430646 | Mar 14, 2013 | Aug 30, 2016 | Fireeye, Inc. | Distributed systems and methods for automatically detecting unknown bots and botnets |
US9432389 | Mar 31, 2014 | Aug 30, 2016 | Fireeye, Inc. | System, apparatus and method for detecting a malicious attack based on static analysis of a multi-flow object |
US9438613 | Mar 30, 2015 | Sep 6, 2016 | Fireeye, Inc. | Dynamic content activation for automated analysis of embedded objects |
US9438622 | Mar 30, 2015 | Sep 6, 2016 | Fireeye, Inc. | Systems and methods for analyzing malicious PDF network content |
US9438623 | Jun 20, 2014 | Sep 6, 2016 | Fireeye, Inc. | Computer exploit detection using heap spray pattern matching |
US20060070048 * | Sep 29, 2004 | Mar 30, 2006 | Avaya Technology Corp. | Code-coverage guided prioritized test generation |
US20060156289 * | Jan 12, 2006 | Jul 13, 2006 | Robert Spinner | Test program set obsolescence mitigation through software and automatic test equipment system processes |
US20070266366 * | Mar 29, 2007 | Nov 15, 2007 | Iosemantics, Llc | Generating and utilizing finite input output models, comparison of semantic models and software quality assurance |
US20080098347 * | Oct 20, 2006 | Apr 24, 2008 | Hana Chockler | Model Checking of Non-Terminating Software Programs |
US20080228679 * | Mar 16, 2007 | Sep 18, 2008 | David Ward | Symbolic depth-first searches using control flow information for improved reachability analysis |
US20090125976 * | Oct 10, 2008 | May 14, 2009 | Docomo Communications Laboratories Usa, Inc. | Automated test input generation for web applications |
US20090172643 * | Dec 23, 2008 | Jul 2, 2009 | Kabushiki Kaisha Toshiba | Program verification apparatus, program verification method, and program storage medium |
US20090228871 * | Mar 10, 2008 | Sep 10, 2009 | Microsoft Corporation | Managing generation of security tests |
US20090265692 * | Oct 22, 2009 | Microsoft Corporation | Active property checking | |
US20090282289 * | May 6, 2008 | Nov 12, 2009 | Microsoft Corporation | Generation and evaluation of test cases for software validation and proofs |
US20100005454 * | Feb 26, 2009 | Jan 7, 2010 | Nec Laboratories America, Inc. | Program verification through symbolic enumeration of control path programs |
US20100100874 * | Oct 22, 2008 | Apr 22, 2010 | Ajikumar Thaitharanikarthu Narayanan | Complexity dependency change impact system and method |
US20100235814 * | Mar 15, 2010 | Sep 16, 2010 | Kabushiki Kaisha Toshiba | Apparatus and a method for generating a test case |
US20100242029 * | Mar 19, 2009 | Sep 23, 2010 | Fujitsu Limited | Environment Data Refinement Based on Static Analysis and Symbolic Execution |
US20110173591 * | Jul 14, 2011 | Target Brands, Inc. | Unit Test Generator | |
US20110185368 * | Jan 26, 2010 | Jul 28, 2011 | Target Brands, Inc. | System and Method for Routing Information |
US20120036499 * | Feb 9, 2012 | Abb Ag | System and method for automatically testing a program for safety-related automation systems | |
US20120110550 * | May 3, 2012 | Indradeep Ghosh | Node computation initialization technique for efficient parallelization of software analysis in a distributed computing environment | |
US20120110591 * | May 3, 2012 | Indradeep Ghosh | Scheduling policy for efficient parallelization of software analysis in a distributed computing environment | |
US20130073837 * | Nov 8, 2012 | Mar 21, 2013 | Concurix Corporation | Input Vector Analysis for Memoization Estimation |
US20140096112 * | Sep 28, 2012 | Apr 3, 2014 | Microsoft Corporation | Identifying execution paths that satisfy reachability queries |
US20140143604 * | Nov 21, 2012 | May 22, 2014 | Fujitsu Limited | Mixed numeric and string constraint analysis |
US20150012729 * | Jul 2, 2013 | Jan 8, 2015 | Arch D. Robison | Method and system of compiling program code into predicated instructions for excution on a processor without a program counter |
USRE43553 * | Jul 24, 2012 | Advanced Testing Technologies, Inc. | Test program set obsolescence mitigation through software and automatic test equipment system processes |
U.S. Classification | 717/124, 717/131, 717/134, 717/127, 717/126 |
International Classification | G06F9/44, G06F11/00 |
Cooperative Classification | G06F11/3676 |
Date | Code | Event | Description |
---|---|---|---|
May 12, 2004 | AS | Assignment | Owner name: MICROSOFT CORPORATION, WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:BALL, THOMAS J.;REEL/FRAME:015339/0772 Effective date: 20040512 |
Sep 14, 2010 | CC | Certificate of correction | |
Apr 15, 2013 | REMI | Maintenance fee reminder mailed | |
Sep 1, 2013 | LAPS | Lapse for failure to pay maintenance fees | |
Oct 22, 2013 | FP | Expired due to failure to pay maintenance fee | Effective date: 20130901 |
Jan 15, 2015 | AS | Assignment | Owner name: MICROSOFT TECHNOLOGY LICENSING, LLC, WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:MICROSOFT CORPORATION;REEL/FRAME:034766/0001 Effective date: 20141014 |